Compare AI models
| Specifications | ||
ParametersThe learned values that shape how a model responds. B means billion and T means trillion. For mixture-of-experts models, the total count is larger than the number active for each token. Parameter count alone does not measure quality. | 1T | 4B |
Context windowHow much information the model can work with in one request, measured in tokens — pieces of words or other data. This includes your question, attachments, conversation and reply. Limits can vary by model variant and provider. | 256k | 128k |
| API pricingUSD per 1M tokens · lower wins | ||
Cheapest inputLowest input rate across every provider on OpenRouter, the lab included. The cheapest endpoint may run a quantised build or a shorter context — see "Available from" on the model page. | $0.60Novita | — |
Cheapest outputLowest output rate across every provider on OpenRouter, the lab included. May come from a different provider than the cheapest input. | $2.50Novita | — |
| $245Novita · bf16 | — | |
| Benchmarks | ||
AIME 2025Reasoning — single-answer accuracy on the 2025 invitational mathematics exam without tools | 94.5% | 46.3% |
GPQA DiamondScience — Graduate-level science questions in biology, physics, and chemistry — hard enough that subject-matter PhDs score around 65%. Higher is better. | 84.5% | 55.1% |
SWE-Bench VerifiedCoding — Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better. | 71.3% | — |
SWE-Bench MultilingualMultilingual coding — Like SWE-Bench, but the coding problems span many programming languages, not just one. Tests how broadly the AI can code. Higher is better. | 61.1% | — |
LiveCodeBenchCompetitive coding — Coding problems published so recently the AI can't have seen them in training — a contamination-free test of raw programming skill. Higher is better. | 83.1% | — |
IFEvalAgentic tasks — following verifiable instructions in generated responses | — | 75.5% |
BrowseCompWeb browsing — Can the AI browse the web and track down hard-to-find answers? Higher is better. | 60.2% | — |
Humanity's Last Exam · text only, no toolsMultidisciplinary reasoning — accuracy on the 2,158 text-only questions without tools | 23.9% | — |
MATH-500Reasoning — accuracy on the 500-problem MATH subset | — | 96.2% |
MMLU-ProKnowledge — accuracy on the ten-option MMLU-Pro benchmark | 84.6% | — |
| Overview | ||
| Company | Moonshot AI | NVIDIA |
| Release date | Nov 6 2025 | May 20 2025 |
| Access | Open Weight | Open Weight |
| Model details | View model | View model |
Other comparisons
Kimi K2 ThinkingvsClaude Fable 5.2Llama Nemotron Nano 4B v1.1vsClaude Fable 5.2Kimi K2 ThinkingvsGPT-6.2 AstraLlama Nemotron Nano 4B v1.1vsGPT-6.2 AstraKimi K2 ThinkingvsGemini 4 NeonLlama Nemotron Nano 4B v1.1vsGemini 4 NeonKimi K2 ThinkingvsMuse EmberLlama Nemotron Nano 4B v1.1vsMuse EmberKimi K2 ThinkingvsGrok 4.7Llama Nemotron Nano 4B v1.1vsGrok 4.7Kimi K2 ThinkingvsDeepSeek-V4.1-Flash-1017Llama Nemotron Nano 4B v1.1vsDeepSeek-V4.1-Flash-1017Frequently asked questions
Kimi K2 Thinking leads Llama Nemotron Nano 4B v1.1 on 2 of the 2 benchmarks they both report (AIME 2025, GPQA Diamond). Llama Nemotron Nano 4B v1.1 shipped 170 days before Kimi K2 Thinking, so benchmark comparisons should account for the intervening progress.
Kimi K2 Thinking has 1T parameters, while Llama Nemotron Nano 4B v1.1 has 4B. Context windows are 256k (Kimi K2 Thinking) vs 128k (Llama Nemotron Nano 4B v1.1).
On AIME 2025, Kimi K2 Thinking leads at 94.5% vs Llama Nemotron Nano 4B v1.1 at 46.3%. On GPQA Diamond, Kimi K2 Thinking leads at 84.5% vs Llama Nemotron Nano 4B v1.1 at 55.1%.